[MPlayer-dev-eng] [PATCH] yuv support for vo_gl

[Reimar Döffinger]

Hi,
On Wed, Sep 14, 2005 at 12:01:54AM +0400, Vladimir Mosgalin wrote:
>  On 2005.09.13 at 18:37:52 +0200, Reimar D?ffinger wrote next:
> > A new version, now also with support for vo_gl2. To get similar speed,
> > you have to use the noglfinish option.
> > Please test extensively if you can, since I intend to apply this version
> > soon.
> 
> It works.
> The problems: there is no OSD at all, no saturation control, brightness
> and contrast are performed in software by X.

Yes, that's "intentional". I don't care much about vo_gl2, and I _hate_
duplicating code. So unless I find a way without duplicating much code I
guess I won't implement it :-).
The OSD can not be done since slice rendering is used, so there is not
really anything I can draw it on. Using -vf expand (with the right
options) is a workaround to get one.

> With noglfinish it's as fast as gl, or maybe a tiny bit faster. Without
> it's 1.4-1.5 times slower. (5% instead of 3% on 640x480, 22% instead of
> 15-16% on 1920x1080)

Btw. with yuv=1 the difference is only minimal, so I'm still quite happy
about implementing it, although it is vendor specific and limited in
functionality :-)

> Actually, I already wanted to suggest you: can you please add the
> ability to load shaders from external files? Either one file with
> separated vertex and fragment codes, or two files. Since many things -

I already thought about that as well as a custom texture fot texture
unit 3, but I am unsure how to add this without making it a hack...
Also - no idea if vertex programs would work right out of the box at the
moment... I think not.

> Of course, supporting GLSlang will make this feature much more useful,
> and I don't even know if it is supported in current drivers, but custom
> files with shader asm will allow to do a lot of interesting effects too.

I don't know. I didn't really like GLSlang from the bit I have been reading
(not to mention that the actual implementations are in a quite unfinished
state it seems).

Greetings,
Reimar Döffinger